Christos Argyropoulos Network Management & Optimal Design Laboratory (NETMODE)

1. PaFloMonA Slice Aware Passive Flow Monitoring Framework for OpenFlow Enabled Experimental Facilities • Christos ArgyropoulosNetwork Management & Optimal Design Laboratory (NETMODE) • National Technical University of Athens • European Workshop on Software Defined Networks • October 26th , 2012 • Darmstadt, Germany

2. Baselineopenflow (OF) enabled experimental facilities • Management Plane • Compute resources provisioning • Flow-space provisioning upon network elements • Access control (authentication & authorization) • Topology creation • Control Plane • Multi-tenant environment with isolation (slicing) • Compute resources isolation • Data plane isolation • Data plane exposure to user’s control logic(via proxy OF controller – e.g. FlowVisor)

3. OFELIA case

4. Is anything missing?

5. Monitoring • Infrastructure Monitoring • Management & control plane services availability • Compute resources state • OpenFlow switches state • FlowVisor state • PHY substrate link state • Virtual Substrate/Slice Monitoring

6. Slice Monitoringtaking into consideration.. • Slice-centric monitoring on heterogeneous network infrastructures • Role-based monitoring data view and easy access for the experimenters • Existence of a monitoring specification that will promote homogeneity and compatibility • Usability (web access & remote calls)

7. OpenFlowfor Slice-based Monitoring?

8. OpenFlow data gathering support • OpenFlow counters • Counters per-table, per-flow, per-port, per queue • Easy fit with the slice-based concept • Slice-based data gathering based on flow-space definition • Proxy Controller permits per-slice counter view to the users’ controllers

9. It seems like a dream!OpenFlow fits everywhere Does it ?

10. OpenFlow constraints • OpenFlow counters • Flow forwarding table polling for statistics is not a good idea (control plane overloading) • You cannot gather flow statistics for a flow you do not have to the forwarding table • Aggregated flow matching during the forwarding process prohibits later on drilling down to micro-flows statistics • Micro-flows use for data gathering (a very bad idea) • OpenFlow forwarding table overflow • High rate of packet-in messages from the OF switches to the OF Controllers (proxy controller bombing)

11. PaFloMon enablers • Taking advantage of existing passive monitoring capabilities • sFlow • Purpose: packet sampling, flow-based exposure to the user • SNMP • Purpose: management info • OpenFlow stats • Purpose: flow-based stats for the applied forwarding rules

12. PaFloMondata gathering on ofelia

13. PaFloMon for Slicestaking into consideration.. • Slice-centric monitoring on heterogeneous networks • Compound wired/wireless (e.g. sFlow on NEC/OpenWRT) • Role-based monitoring data view and easy access for the experimenters • Roles correspond to slice user-view and admin-view • Monitoring spec for homogeneity and compatibility • Monitoring RSpec for integration of monitoring requests • Usability (web access & remote calls) • Web-based and XML-RPC access of the stored data

14. PaFloMonhigh level overview

15. PaFloMonfeasibility tests on ofelia gear • Micro-flows to the NEC switch • Rates that lead to tables’ overflow affect both packet forwarding & sFlow sampling • “flow Number of flow entries per slice per dpid (switch)” parameter • Must be carefully configured to avoid performance degradation of the entire infrastructure • The sum of the maximum numbers of flows per slice per switch, permitted by the FlowVisor, should not exceed the switch flow-table size limit (Σflowsmax, slice κ, switch λ≤ table_sizeswitchλ) κ

16. Thank You!